Machine Learning Models for Blood Glucose Prediction in Diabetes Management PI: Cindy Marling Co-PIs: Razvan Bunescu and Frank Schwartz, MD This illustration, provided by Medtronic, shows a patient wearing: A) an insulin pump (essentially a computer-controlled insulin syringe); B) tubing that takes insulin from the pump to an infusion set (shown above the tubing); C) a glucose sensor; and D) a transmitter that sends data from the sensor to the pump for storage. The Problem Nearly two million Americans have type 1 diabetes, a chronic disease in which the pancreas does not produce insulin. Type 1 diabetes is treated with insulin therapy and managed through blood glucose control. Good blood glucose control is essential for patients to avoid serious diabetic complications, including blindness, amputations, kidney disease, strokes, heart attacks, and death from severe hypoglycemia. Achieving and maintaining good blood glucose control is difficult. Patients are highly individual in their responses to treatment and to life events that impact blood glucose levels. Large volumes of blood glucose data may be collected automatically by continuous glucose monitoring systems (see illustration at lower right), but automated analysis is lacking. Patients do not always know when problems are impending; problems occurring while patients are asleep are especially dangerous. The Goal Advances in computing technology that account for individual differences and exploit underutilized data could improve diabetes management and promote personalized medicine. Accurately predicting impending blood glucose control problems would enable preemptive intervention, leading to improved overall control. Machine learning models that predict blood glucose levels would enable or facilitate new applications of direct benefit to patients, including: alerts to immediately notify patients of imminent problems; decision support systems recommending actions to prevent problems; and educational simulations showing the effects of different treatment choices or lifestyle options on blood glucose levels. The Science The task of blood glucose prediction is approached as a time series forecasting problem. Support vector regression models are being investigated for predicting blood glucose levels based on a patient’s prior blood glucose levels, insulin data, meal data, exercise data, sleep patterns and work schedules. (See overview The Science (continued) figure above.) To account for individual patient differences, separate models are trained for each patient. Additionally, transfer learning may enable data from multiple patients to aid in building models for patients with limited historical data. Models are sought that are robust in the face of imperfect data, including missing life events, inaccurately recorded life events, and noisy glucose sensors. Models are tailored to provide best performance in domain dependent critical situations, such as impending hypoglycemia. The Broader Impact This work strives to improve the overall health and quality of life for people with type 1 diabetes. The most significant impact could come from the ability to predict nocturnal hypoglycemia to combat the “dead in bed” syndrome. In addition, a new SmartHealth Lab was founded at Ohio University to support this work, promote additional interdisciplinary smart health research, and attract more women to careers in computer science. Acknowledgements This material is based upon work supported by the National Science Foundation under Grant No. IIS We gratefully acknowledge the collaboration of physicians Jay Shubrook, DO, and Aili Guo, MD, PhD, and the contributions of graduate research assistants Melih Altun, Michael Cleaver, Nattada Nimsuwan, and Nigel Struble. Overview of the Blood Glucose Prediction Process